Intro
Clément Bourgeois

Clément Bourgeois

Development Team Manager · Engineering & Delivery · Software Craftsmanship & AI Explorer

I build teams that deliver.

  • Engineering & Delivery
  • Software Craftsmanship
  • AI Explorer
  • Human Leadership
  • Systems Thinking

And, in my spare time, creator of tech projects.

I help teams navigate technological transformations, from agility to software craftsmanship and now the shift toward artificial intelligence, by combining human leadership, delivery mindset, creativity, and a culture of continuous improvement.

My conviction is simple: there is only one constant in the universe, and that is change. My role, within the organization, is to help teams not endure it, but turn it into leverage.

My role sits at the intersection of people management, delivery management, and software engineering. I support teams in collaborating better, building better, and delivering better, with the belief that delivering together creates more value than delivering separately.

I am particularly interested in Software Craftsmanship, software quality, automation, and the impact of AI on the way we develop. As artificial intelligence progressively transforms our professions, I see a strong opportunity to strengthen our practices, learn faster, and help teams and organizations evolve sustainably.

This site is a personal showcase of my background, technical projects, and experiments around software, AI, automation, and agents.

Links

Bio

I am a technology professional with a background at the crossroads of software development, technical leadership, and team management.

Over the years, I have supported development teams in designing, delivering, and improving complex software solutions. My approach combines a strong sensitivity to code quality, architecture, modern engineering practices, and the ability of teams to improve sustainably over time.

I am particularly interested in software craftsmanship, automation, AI applied to development, autonomous agents, tools that increase team productivity, and practical ways to turn ideas into useful solutions.

This site brings together some of my personal and technical projects, experiments, and the tools I build to learn, test, automate, or simply push an idea one step further.

How I work

I help teams move through change by keeping people, quality, and human responsibility at the center.

Successful transformation starts with a clear vision, genuine listening, and an honest understanding of what is blocking progress. I help teams build a concrete action plan grounded in autonomy, ownership, accountability, and sustainable quality.

  1. Create meaning

    Explain the vision, clarify the why, and make objectives understandable for everyone.

  2. Listen to the field

    Understand people, friction points, and blockers before proposing a solution.

  3. Build with the teams

    Create an action plan with the teams at the center, encouraging autonomy, ownership, and accountability.

  4. Keep quality at the center

    Use communication, collaboration, transparency, and traceability to deliver with confidence.

  5. Use AI responsibly

    Make AI an accelerator for understanding, quality, and progress, without transferring the human responsibility for delivery.

Learn more about my approach

I deeply believe that people who respect each other, communicate well, and understand their strengths as well as their limits become stronger together than the sum of their individual contributions.

Communication is the backbone of teamwork, but it only works when it is grounded in genuine respect for people. This human foundation is what allows a team to move through change without losing cohesion.

AI is transforming our professions and becoming a powerful accelerator to better understand, build, and improve quality. But AI is not responsible for delivery. That responsibility remains human, and it should remain human.

Projects

This section presents a few technical projects I am currently building or evolving. Some are personal experiments, others are more structured tools, but they all share the same idea: learning by building.

Agent Sommelier

Public · Live

An AI sommelier that models taste to recommend wines based on your profile, your friends' preferences, and the meal you are sharing.

The project combines sensory profiles, embeddings, vector search, and LLM reasoning to produce personalized recommendations. Each tasting refines an 8-dimensional taste profile, which is used to compare preferences, recommend bottles, and find multi-guest wine pairings around a dish.

Tech stack:

  • Next.js
  • FastAPI
  • PostgreSQL
  • pgvector
  • Redis
  • Claude
  • Voyage AI
  • Docker
  • Raspberry Pi
  • Playwright

System logic

  1. Tastings
  2. Sensory profile
  3. Vector search
  4. Justified recommendation
Architecture & engineering highlights

Agent Sommelier is built around a taste-oriented RAG architecture. Bottles, tasting notes, and user profiles are projected into the same vector space with Voyage AI, then stored in PostgreSQL with pgvector. This allows semantic search to run close to the business data without relying on a separate vector database.

Its signature feature is multi-guest wine pairing: the dish and guest profiles are combined to retrieve candidate bottles, then an LLM re-ranks them into justified recommendations. The system intentionally separates catalog-constrained recommendations from open suggestions, making the distinction between controlled retrieval and free generation explicit.

From an engineering perspective, the product runs in production with a FastAPI backend, a Next.js frontend, Redis workers, PostgreSQL 16, Docker Compose, Traefik, and Cloudflare Tunnel on a Raspberry Pi 5. It also includes LLM cost observability, granular permissions, rate limiting, Playwright tests, and an AI-agent-assisted development workflow.

Visit Agent Sommelier

Agent Voyage Intelligent

Public · Live

A travel planning assistant that turns your ideas, preferences, and constraints into a structured, realistic, and easy-to-follow itinerary.

The project covers end-to-end planning: optimized itineraries adapted to your pace, grouping activities by zone to minimize travel, smart alternatives for rain or unexpected events, checklists generated for each phase of the trip, daily maps, and budget and booking tracking.

Tech stack:

  • Next.js
  • React
  • TypeScript
  • FastAPI
  • PostgreSQL
  • SQLAlchemy
  • Pydantic
  • JWT
  • dnd-kit
  • Google Maps
  • Docker
  • Traefik
  • Cloudflare Tunnel
  • Raspberry Pi

System logic

  1. Trip description
  2. Structured itinerary
  3. Zone optimization
  4. Maps & checklists
Architecture & engineering highlights

Agent Voyage is built on a decoupled architecture: a Next.js 14 (App Router) frontend in React and TypeScript, and an async FastAPI backend in Python 3.12. Persistence relies on PostgreSQL 16 with SQLAlchemy 2.0 and Alembic migrations, with data validation handled by Pydantic v2.

Authentication is built in-house with JWT tokens (python-jose) and bcrypt-hashed passwords. Itinerary reordering is done via drag & drop (dnd-kit) and locations are displayed on Google Maps. Abstract interfaces are designed to plug in external connectors and LLMs later on.

On the infrastructure side, everything is containerized with Docker Compose (frontend, backend, PostgreSQL) and self-hosted on a Raspberry Pi. Traefik handles reverse proxying and label-based routing, while Cloudflare Tunnel exposes the site over HTTPS without opening a single port — with separate Docker networks to isolate the database, health checks, rate limiting, and secure headers.

Visit Agent Voyage Intelligent

Agent Recette

Public · Live

A personal assistant for cooking and cocktail recipes that gathers, structures, and organizes your recipes from any source.

The project covers the recipe library end to end: manual entry, web import (schema.org/Recipe with a Claude fallback), and extraction from a video (YouTube, Facebook, Instagram). Browse with serving scaling and a cooking mode, choose metric or imperial units, and build a persistent shopping list that aggregates ingredients by aisle with smart quantity merging.

Tech stack:

  • Next.js
  • React
  • TypeScript
  • FastAPI
  • PostgreSQL
  • SQLAlchemy
  • Alembic
  • Pydantic
  • JWT
  • Claude
  • Whisper
  • yt-dlp
  • Docker
  • Traefik
  • Cloudflare Tunnel
  • Raspberry Pi

System logic

  1. Source (entry, web, video)
  2. Extraction & structuring
  3. Normalized recipe
  4. Library & shopping list
Architecture & engineering highlights

Agent Recette is built on a decoupled architecture: a Next.js 14 (App Router) frontend in React and TypeScript, an async FastAPI backend, a dedicated PostgreSQL 16 database with SQLAlchemy 2.0 and Alembic migrations, and a separate transcription worker. Authentication is built in-house with JWT tokens and bcrypt-hashed passwords.

Sources are treated as connectors plugged around a common core: the structured recipe. Web import leverages schema.org/Recipe markup with a Claude fallback; video import chains the description, an optional recipe link, and subtitles (via yt-dlp), then, as a last resort, transcribes the video's audio with a local Whisper model (VAD in a dedicated worker). Everything is structured by Claude under one strict rule: never invent a quantity.

On the infrastructure side, everything is containerized with Docker Compose (frontend, backend, dedicated PostgreSQL, transcriber) and self-hosted on a Raspberry Pi. Traefik handles reverse proxying and label-based routing, while Cloudflare Tunnel exposes the site over HTTPS without opening a single port, with separate Docker networks to isolate the database.

Visit Agent Recette

Living Legacy

Public · Live

A web app designed to immortalize the memory of a loved one — often elderly. The person tells their life out loud, at their own pace; each story is transcribed word for word and kept exactly as told, as a faithful, unalterable trace of their voice.

But the real value comes from consolidation: these stories, scattered in every direction, are reorganized into a structured, navigable life book, enriched with each recording and without ever inventing anything. To this are added the family tree and the circle of relatives, friends, and acquaintances who mattered — preserving not only the memories, but the bonds of a whole life.

Everything stays private by default and shareable with the people dear to them. Bilingual (FR / EN) and designed to be accessible to an older audience.

Tech stack:

  • FastAPI
  • Next.js
  • React
  • PostgreSQL
  • Whisper
  • Claude
  • audio transcription
  • family tree
  • Docker
  • Traefik
  • Cloudflare Tunnel
  • Raspberry Pi

System logic

  1. Tell out loud
  2. Word-for-word verbatim
  3. Book, tree & circle
  4. Read & share
Visit Living Legacy

NAS100 Trading Agent

Private

An autonomous algorithmic trading agent designed to operate in real time on the Nasdaq 100.

The system covers the full quantitative trading lifecycle: historical backtesting, automated parameter optimization, live execution, and continuous monitoring. Its core strength is a strict equivalence between simulation and production: the same decision logic drives both testing and live trading, reducing the gap between theoretical performance and real market behavior.

The agent includes market regime detection, risk management, intelligent exit rules, and robustness-oriented scoring. It runs on a lightweight automated infrastructure, deployed with Docker on a Raspberry Pi.

Technologies and topics explored:

  • Python
  • algorithmic trading
  • backtesting
  • optimization
  • risk management
  • Docker
  • Raspberry Pi
  • monitoring

System logic

  1. Market data
  2. Strategy logic
  3. Live execution
  4. Monitoring

Home Automation Agent

Private

A project focused on home automation and interaction with connected devices.

The goal is to create an agent capable of understanding simple requests, interacting with the home environment, triggering actions, and eventually offering a more natural way to control the house.

Technologies and topics explored:

  • Home automation
  • connected devices
  • APIs
  • Raspberry Pi
  • AI assistants
  • service orchestration

System logic

  1. Home context
  2. Intent understanding
  3. Device action
  4. Confirmation

Digital Twin House Agent

Coming soon

A future project aiming to create a digital representation of the house, its equipment, spaces, and behaviors.

The goal would be to build an agent capable of understanding the state of the house, simulating certain scenarios, anticipating needs, and helping manage the home environment more effectively day to day.

Technologies and topics considered:

  • Digital twin
  • home automation
  • sensor data
  • visualization
  • automation
  • simulation
  • contextual AI

System logic

  1. Home data
  2. Simulation
  3. Insight
  4. Automation
Featured

Agent Sommelier

An agent dedicated to taste discovery, wine and spirits recommendations, and food pairing suggestions.

The goal is to help someone better understand their preferences, discover new bottles, and receive suggestions adapted to their taste profile, context, and meal.

Open Agent Sommelier

Publications

A few press mentions and appearances.

What I am interested in

I am interested in projects where technology makes things simpler, smarter, or more useful.

My main areas of interest include software development, architecture, automation, artificial intelligence, autonomous agents, software quality, and the practices that help teams build better things together.

This site is both a showcase, an experimentation space, and a technical logbook.